[beinan]

Summary

Logical checkpointing leaves obsolete native log segments on disk. Add an explicit node-wide maintenance handle that rewrites retained entries and metadata before reclaiming those segments, plus configurable rotation and purge thresholds.

Maintenance drains writes through a separate gate and syncs a nonempty marker before native reclamation; uncertain errors fence every group sharing the engine. Ordinary reads remain concurrent with ordinary fsyncs. The native maintenance pass temporarily excludes payload reads and persistence and has no per-call byte or time limit; scheduling, admission quotas and application snapshot collection remain separate.

Refs #378.

Testing

• All-feature WAL unit tests: 105 passed, 1 ignored.
• Actual TCP/Lance process tests: 10 passed, 1 child fixture ignored; native segments are removed before lagging-replica snapshot transfer, interrupted transfer and all-replica restart.
• Repeated real-file append/checkpoint/purge tests verify removed paths, retained suffixes and an uncheckpointed sibling after reopen; injected rewrite-sync failure fences every shared group. A blocked ordinary fsync does not block a sibling durable payload read.
• Feature Clippy and no-default compilation passed.
• Full repository helper: 628 Rust tests and 229 Python tests passed (2 skipped, 1 expected failure), plus formatting, workspace Clippy, Ruff and Pyright.
• Workspace packaging with all features and final source/binary integrity checks passed.